What an ATS actually parses, and the five formatting choices that break it

Half of what you have read about applicant tracking systems is fear and the other half is folklore. So let me describe what the software actually does with your file. Once you have seen it, the formatting rules stop feeling like superstition and start looking like plumbing.

An ATS has one job at the moment you press submit: turn a document into rows in a database. Name. Email. Phone. A list of employments, each with a company, a title, a start date and an end date. A list of qualifications. A bag of skill keywords. Workday, Greenhouse, Lever, Taleo, SmartRecruiters and the parser sitting behind most job-board uploads all do the same thing. None of them gives your CV a mark out of a hundred and rejects you at 74. That number does not exist. What exists is a search box, and a recruiter typing "IFRS" AND "SAP" into it at nine in the morning with 340 applications to get through before lunch.

So the real risk is not rejection by a machine. The real risk is being unfindable, or being parsed into nonsense that a human would have to repair by hand and never will.

What actually happens to your PDF

The parser flattens your file into a stream of characters with coordinates, then infers structure from three signals: headings it recognises, date patterns, and reading order. Reading order is the one nobody thinks about. A PDF does not store "column one, then column two". It stores glyphs at positions, and the parser guesses the sequence. Guess wrong and your job titles get welded to somebody else's dates.

The five choices that break it

1. Two columns

This is the big one, and it is the default in most of the pretty templates being sold online. A left sidebar holding skills, languages and contact details, with experience running down the right, looks superb on screen. When the parser reads it line by line across the page, you get output like Cairo University Advanced Excel 2016 Arabic (native) Senior Accountant. Every field is present. None of them belongs to anything.

Some modern parsers handle two columns. You do not know which one is on the other end, and you have one attempt. Use a single column for anything you upload to a portal. Keep the two-column design for the version you email to a human or hand over in an interview.

2. Text living inside images

A name rendered as a logo, a skills chart drawn as a graphic, contact details inside a coloured banner image: all invisible. There is no OCR step in a CV parser. If a piece of information exists only as pixels, it does not exist. The same goes for icon-only contact rows — a phone glyph next to a number tells the parser nothing, and the word "Phone" costs you five characters.

3. Headers and footers

Word and InDesign header/footer regions frequently do not make it into the extracted text at all, or arrive detached at the very end of the stream. Putting your name, email and phone in the page header is the single most self-defeating thing I see, because the one field the system needs above all others is the one you hid. Contact details belong in the body of page one, as plain text.

4. Tables holding the experience section

Tables are fine for a short skills grid and dangerous for employment history. Merged cells and nested tables are where parsers reverse row order or drop a cell entirely. If your dates sit in a narrow left column and your descriptions in a wide right one, you have rebuilt problem number one with extra steps.

5. Invented section headings

"My Journey", "Where I've Made Noise", "Toolbox". The parser is matching against a dictionary of known headings. Experience, Work Experience, Education, Skills, Certifications, Languages. Be boring in the headings and interesting in the sentences underneath them. Nobody was ever hired for a clever section label.

The things that genuinely help

• Dates in one consistent format, written as Mar 2021 – Aug 2024. All-numeric formats invite a parser set to US conventions to read 03/04/2021 as 4 March when you meant 3 April. Write "Present" for current roles, never "Till date" or "Now".
• Company, then title, then dates, in that order, on their own lines. One employment block per job, never two roles merged because they were at the same company. Promotions get their own block; that is how a parser sees a promotion at all.
• A real file name. Mona-Adel-Financial-Analyst-CV.pdf gets forwarded. cv final final2 (1).pdf gets a raised eyebrow before it is opened.
• PDF exported from a text tool, not a scan and not an exported image. If you cannot select and copy a sentence out of your own PDF, no parser can read it either.
• Both spellings of everything. Write "Search Engine Optimisation (SEO)" once, in full and abbreviated. Recruiters search for one or the other, rarely both.

The keyword question, answered honestly

Keyword stuffing works about as well as it did on web pages in 2009. White text on a white background is detected by every serious parser and by every recruiter who ever pressed Ctrl+A. The version that does work is unglamorous: read the posting, list the eight to ten nouns it repeats, and make sure every one of them appears somewhere true in your CV.

If the posting says "revenue recognition" and your CV says "sales accounting", you are describing the same work in a vocabulary the search box will not find. Change the words, not the facts.

The bar is not "beat the robot". The bar is "be searchable, be readable, and survive being converted to plain text".

A five-minute test before you apply

1. Open your PDF. Select all, copy, paste into a blank plain-text document.
2. Read what comes out. If the order is wrong, your layout is wrong.
3. Check your phone number and email survived, in one piece, near the top.
4. Check every employer, title and date pair is still adjacent to the others in its block.
5. Search the plain text for the five most important nouns from the job posting.

If that text version reads sensibly, the parser will be fine and so will the human. Most CVs that fail this test fail it spectacularly, and the person who wrote them never finds out — which is exactly why it is worth the five minutes.